chunked/content_aware_chunking/_fsdp/chunk_14.txt

The wrapping is usually applied in a nested way where the full weights are discarded after each forward pass to save memory for use in the next layer.